Recursive Unveils Flint, a Model Built by Its Own AI System
Recursive's first release demonstrates a pipeline where the AI designs its own architecture and training methods.
By Anastasia Volkova ·
The fear that artificial intelligence could eventually design itself into something unstoppable is a common talking point for critics. However, Recursive is attempting to prove that this capability, known as recursive self-improvement, is not just a hazard but a practical way to build better software right now. The company has released Flint, an open-weight coding model that was largely constructed by an AI system designed to automate the creation of other AI.
According to Richard Socher, Recursive's CEO and cofounder, the company relied on an internal system called the Eureka Machine to handle nearly every aspect of the model pipeline. This automation allowed a team consisting of only a handful of researchers to produce a competitive model in a matter of months. The Eureka Machine generated a new way to train the base model, invented new model architectures, and produced code optimized for specific silicon chips. Recursive shared benchmark scores indicating that Flint outperforms other models of similar size and architecture on coding tasks.
Recursive has raised $650 million at a $4.65 billion valuation from investors including Google Ventures, Greycroft, Nvidia, and AMD Ventures. Socher was previously the chief scientist at Salesforce, and his cofounder, Jeff Clune, is a professor at the University of British Columbia who researches methods for building AI systems capable of learning indefinitely.
The stakes for this technology are high. Some experts warn that allowing AI to improve itself indefinitely could lead to disaster, with concerns about human extinction gaining attention after a researcher left Anthropic in September to warn of such risks. Several bills in Congress have even proposed to ban recursive self-improvement. Recursive views these fears as more science fiction than reality, arguing that legislation to prevent such research is an unnecessary impediment to progress.
Socher insists that Flint was built using a system incapable of escaping human control. While the goal is for the Eureka Machine to eventually achieve full recursive self-improvement, Socher says it would still be overseen by humans who could stop the process if it goes awry. "We hope that this shows people that indeed we have all the steps in place" to build such a system, he says.
Flint is a 40 billion parameter model, a size that can run on a high-end consumer machine rather than requiring racks of industrial servers. Testing the model for coding tasks, including building command line tools and websites, showed it to be fast and capable without the errors often found in models of this size. While other labs are also pursuing recursive self-improvement, Recursive believes this approach is the only way a small team can produce a model that outperforms others in its class.
Source: Wired
